Computing Graph Descriptors on Edge Streams

Feature extraction is an essential task in graph analytics. These feature vectors, called descriptors, are used downstream vector-space-based analysis models. This idea has proved fruitful the past, with spectral-based descriptors providing state-of-the-art classification accuracy. However, known algorithms to compute meaningful do not scale large graphs since: (1) they require storing entire memory, and (2) end-user no control over algorithm’s runtime. In this article, we present streaming approximately three different capturing structure of graphs. Operating on edge streams allows us avoid controlling sample size enables keep runtime our within desired bounds. We demonstrate efficacy proposed by analyzing approximation error Our scalable millions edges minutes. Moreover, these yield predictive accuracy comparable methods but can be computed using only 25% as much memory.